School and training centre

What the platform does in a training centre and what that is based on

What this actually is

There is no «School» industry template in this build: it needs a class schedule module, and that has not been built. The dental appointment machinery does not fit — there one patient occupies a slot; here there is a group, attendance and a substitute teacher.

The work runs on general modules configured for a training centre, and we say so plainly: a «for schools» page built on general modules is a «for anybody» page with a heading about schools, unless it says which part of it is industry-specific.

Which modules are switched on

The set the internal pilot ran on. Every module is switched on and off separately — this is a first composition, not a decision that would later have to be undone whole.

• Knowledge base: confirmed answers about prices, schedule, enrolment, refunds and discounts.
• Support: parent requests with a queue, a deadline and handover to a person.
• Sales: website requests as leads, the enrolment funnel.
• Finance: the enrolment budget, invoices for approval.
• Analytics: monthly series with a recorded source for every number.

Which AI employees work here

All fifteen were created in the internal pilot; one was moved into work by the decision of two people. For a training centre these are the ones with practical meaning:

• Support — the subject of a request and a drafted reply from the confirmed knowledge base.
• Knowledge — outdated answers and answers that contradict each other.
• Sales — sorting an incoming request about a course.
• Finance — reconciling the enrolment plan with the fact.
• BI analyst — explaining a change in a figure with the source of every number.

What this is based on

The page rests on the internal pilot «AAA AI School», run on 2026-09-16. That is a PILOT OF THE PRODUCT on training centre data, not a deployment at a customer: there has not been a single customer pilot, and we will not present an internal run as somebody else’s result.

What the run produced: 46 knowledge base answers were created, all confirmed by a second person. Of thirteen parent questions, ten were answered from the base with sources cited, and three questions outside the base got an honest «no data». Zero questions were lost: every question and every answer is a row in the database.

On language specifically: the built-in provider matches the words of the question against the words of confirmed answers, so a question in Kazakh against a base in Russian matches nothing — in the run it honestly got «no data». Answering on the merits in Kazakh and English is what an external model does, and it is connected by a key.

What is not in this build

The internal pilot ran on the built-in deterministic provider: not one answer came from an external model, and outbound calls were zero. That is an honest acceptance test of the product and not a measurement of an external model.

• Class schedule, attendance and teacher substitution — there is no module.
• Hours saved: computing them from a run on pilot data would present a guess as a measurement.
• Student and parent satisfaction: there were no live users and no survey was run.
• Sending messages to parents: the platform writes to nobody on its own.

What is the same in every industry

Company isolation enforced by the database, a closed register of permissions, an immutable journal, and a second person signing off where the action calls for it. That is how the platform is built, and it does not depend on which modules are switched on.

In every industry an AI employee produces a draft. Sending a message, paying, signing and deciding about a person are things the platform does not do — in no mode and under no role.

Basis: docs/SCHOOL_PILOT_REPORT.md, packages/contracts/src/project.ts. This page describes what exists in this build, not what is planned.